Research map / Reinforcement learning / Offline RL
Flow to Control: Offline Reinforcement Learning with Lossless Primitive Discovery
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use lossless invertible primitives instead of IQL's direct action policy | IQL · Antmaze-medium-play | normalized score: 71.2 vs 74.7±2.2 | Helped |
| use lossless invertible primitives instead of IQL's direct action policy | IQL · Antmaze-large-play | normalized score: 39.6 vs 56.2±3.6 | Helped |
| use lossless invertible primitives instead of IQL's direct action policy | IQL · Antmaze-medium-diverse | normalized score: 70.0 vs 83.7±1.6 | Helped |
| use lossless invertible primitives instead of IQL's direct action policy | IQL · Kitchen-complete | normalized score: 62.5 vs 75.0±0.7 | Helped |
| use lossless invertible primitives instead of IQL's direct action policy | IQL · Adroit-pen-human | normalized score: 71.5 vs 63.1±1.6 | Hurt |
| use lossless invertible primitives instead of IQL's direct action policy | IQL · Adroit-hammer-human | normalized score: 4.4 vs 3.3±0.7 | Hurt |
| replace VAE-style OPAL primitives with invertible LPD primitives | IQL+OPAL · Antmaze-medium-play | normalized score: 48.6±1.0 vs 74.7±2.2 | Helped |
| replace VAE-style OPAL primitives with invertible LPD primitives | IQL+OPAL · Kitchen-complete | normalized score: 11.5±2.0 vs 75.0±0.7 | Helped |
| replace VAE-style OPAL primitives with invertible LPD primitives | IQL+OPAL · Kitchen-partial | normalized score: 74.9±0.3 vs 72.5±1.1 | Hurt |
| replace VAE-style OPAL primitives with invertible LPD primitives | IQL+OPAL · Antmaze-large-play | normalized score: 56.9±3.3 vs 56.2±3.6 | No clear effect |
| use temporally extended primitives rather than one-level control (ablation) | IQL+LPD · Kitchen-partial-v0 | normalized score | Helped |
| set skill length c to around 5 (ablation) | IQL+LPD · Kitchen-partial-v0 | normalized score | Helped |
Closest papers by meaning
- OPAL: Offline Primitive Discovery for Accelerating Offline Reinforcement Learning (ICLR 2021 · similarity 0.91)
- Offline Hierarchical Reinforcement Learning via Inverse Optimization (ICLR 2025 · similarity 0.91)
- Model-Based Offline Planning (ICLR 2021 · similarity 0.89)
- Offline RL with Observation Histories: Analyzing and Improving Sample Complexity (ICLR 2024 · similarity 0.88)
- Learning from Reward-Free Offline Data: A Case for Planning with Latent Dynamics Models (NeurIPS 2025 · similarity 0.88)
- Are Expressive Models Truly Necessary for Offline RL? (AAAI 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.